Boards are hearing a consistent message from their executive teams: AI is in use, investment is rising and the productivity payoff is close. The evidence beneath that message is less tidy. Surveys of executives, employees and firms show a recurring pattern. Leadership belief in AI’s progress is running ahead of the organization’s capacity to execute. This AI confidence gap is not a communications problem or a temporary lag. It shapes how capital is allocated, how workforce commitments are made and how risk is governed. It should be treated as a strategic exposure in its own right.
The stakes rise with the spend. A National Bureau of Economic Research working paper reports that 78% of firms use AI, with US firms expecting to spend $2,000 per employee on AI-related expenditures in 2026, implying around $300 billion in aggregate. When budgets of that size rest on expectations rather than demonstrated capability, the cost of being wrong compounds.
What the evidence says about the AI confidence gap
Adoption depends on who is asked
Estimates of how widely AI has been adopted vary sharply by source. Census Bureau business survey data show that about 18 percent of firms had adopted AI as of year-end 2025. That is well below the figures that executive surveys produce. Methods differ, so the numbers are not strictly comparable. Still, the divergence is instructive. The closer a measure sits to operational reality, the less advanced adoption tends to look.
Spending is also far from even. The same NBER research finds that just 2% of firms account for over 50% of total AI spending. For the typical enterprise, executive confidence is often benchmarked against a small group of heavy investors. Those firms’ budgets, data estates and talent pools may look nothing like its own.
Expected returns run ahead of realized ones
Executives report that AI adoption has raised annual labor productivity growth by 0.2% over the last three years, but expect a 0.6% per annum boost over the next three. A substantially larger effect may prove right. But it is a forecast, and forecasts tend to be written into budgets as commitments.
A separate NBER study of corporate executives documents a productivity paradox, in which perceived productivity gains are larger than measured productivity gains, likely reflecting a delay in revenue realizations. The benign reading is that value is in the pipeline. The less comfortable reading is that leaders are scoring themselves on perceived gains while measured gains have yet to appear, and that boards are being asked to fund the difference.
The view from the frontline
Employee data show where the gap is widest. In Gallup’s 2025 data, 44% of employees say their organization has begun integrating AI, but only 22% say it has communicated a clear plan or strategy for doing so. Reported frequent AI use by production and front-line workers was essentially flat, moving from 11% in 2023 to 9% in 2025.
Usage also tracks hierarchy. Among organizations that provide AI tools, 67% of leaders report using AI frequently, compared with 46% of individual contributors. Leaders who use AI regularly and find it useful naturally extrapolate that experience across the organization. That extrapolation is how the confidence gap forms. The people setting expectations are the people least likely to meet the friction that stalls adoption further down.
The experience of use is also more muted than the narrative suggests. Among weekly users, only 17% give the highest possible rating to AI’s impact on their productivity and efficiency.
Why overconfidence is a strategic risk, not a sentiment
It is tempting to read executive optimism as a healthy sign of commitment, and optimism does mobilize sponsorship. But when it is not tied to evidence of operating readiness, it creates four distinct exposures.
- Capital misallocation. Organizations fund scaled rollouts before workflows, data and controls are ready. They book benefits that arrive late or not at all. When sponsors lose patience, programs are cut abruptly. This is a familiar pattern behind why enterprise AI strategies fail to produce operating results.
- Workforce commitments built on unverified assumptions. The NBER evidence shows that more than 40% of firms anticipate positive or negative employment effects that roughly offset only in the aggregate. It also shows that larger companies anticipate AI-driven workforce reductions, while smaller firms expect modest gains. Reductions taken ahead of proven capability can remove the people who understand how the work is actually done.
- Governance blind spots. Confidence in sanctioned programs can hide what is happening outside them. The Federal Reserve observes that unsanctioned shadow AI use is unlikely to still be a major driver of differences in adoption estimates, but it remains a substantive problem for organizations. Closing that gap takes controls a risk function can actually test, not policy statements alone.
- Credibility erosion. When forecasts slip, boards discount the next forecast. A full reset of a program that has lost sponsor confidence costs more than calibrating expectations early.
Diagnosing the gap: four dimensions that matter
Surveys measure what executives believe. A useful diagnostic measures whether the organization can act on that belief. Four dimensions account for much of the distance.
Decision rights
AI programs often sit with the technology function while the value they promise sits with business units. When that happens, nobody owns the outcome. The diagnostic question is simple. For each priority use case, can leadership name the business owner accountable for the benefit? Can it name who may approve a change to the workflow, and who may pause an AI system that misbehaves? Ambiguity here is an early sign that confidence has outrun structure.
Process maturity
AI amplifies the process it is placed in. Gallup finds that 88% of employees who strongly agree that AI integrates well with the systems and processes they use at work use AI frequently, compared with 55% of those who do not strongly agree. Integration, not enthusiasm, drives sustained use. Leaders should ask whether target workflows are documented, stable and measured, and whether data is reliable at the point of use. A use case grafted onto an unstable process will disappoint, however capable the model.
Incentives
If managers are rewarded for throughput under existing methods, adopting a new tool is a short-term cost with an uncertain payoff. Performance measures, budget rules and headcount planning send louder signals than strategy documents. Gallup’s engagement research makes a related point: making AI available is not the same as having a clear AI implementation plan. Aligning performance, reward and workforce planning with AI objectives is often the step that turns stated priorities into changed behavior.
Talent
The talent question is broader than hiring data scientists. It covers managers who can redesign work, frontline staff who can supervise and correct AI output, and risk teams who can test controls. Flat frontline usage suggests that capability building has not reached the part of the organization where much of the operational work happens. That is a change management and adoption challenge as much as a technical one.
Strategic and operational implications for enterprise leaders
Treating overconfidence as a managed risk changes how AI programs are governed, funded and reported. Six practical shifts follow.
- Separate belief from evidence in board reporting. Report AI progress in two columns, expected value and measured value, and track whether they converge. The productivity paradox makes this distinction essential rather than pedantic.
- Measure readiness where work happens. Enterprise-wide averages hide the functions that will stall. A structured AI readiness assessment at function level gives leaders a defensible baseline before committing capital.
- Sequence investment to readiness. Fund deployment waves where decision rights, processes and data are ready. Treat other areas as candidates for remediation, not deployment. The tradeoff is a slower headline adoption number in exchange for lower write-off risk.
- Make workforce plans conditional. Tie redeployment or reduction decisions to demonstrated capability milestones, not forecasts. This keeps the option to adjust if expected gains arrive later than planned.
- Assign an owner for calibration. Give a named executive or board committee responsibility for challenging AI forecasts, as would be expected for any other major capital program.
- Communicate the plan, not just the tools. The gap between employees who see AI being integrated and those who have heard a clear plan is itself a readiness risk. It is also one of the cheapest to close.
Conclusion: closing the AI confidence gap
The AI confidence gap is not proof that AI is overhyped or that executives are wrong about direction. Their expectations may prove correct. The issue is timing and grounding. Belief that is not tested against operating evidence leads to premature spend, premature workforce decisions and governance that lags real usage. Organizations that treat confidence as an input to be verified will make steadier progress than those that treat it as a result to be celebrated. When the board next hears that AI is on track, the right questions are: on track according to whom, and measured how?
Where WGA Consulting fits
WGA Consulting’s AI Workforce Readiness Assessment is a fixed-fee diagnostic. It benchmarks an organization across eight domains using WGA’s RCP v2 framework of 40 questions and five maturity levels. The output includes maturity scoring by domain, a function-level readiness heat map, friction and value mapping, and an executive decision readout. It also produces a sequenced roadmap for deploying AI agents in waves. For leadership teams concerned about an AI confidence gap, this offers a practical way to test the decision-rights, process, incentive and talent questions above against evidence rather than sentiment.
Where a program has already lost sponsor confidence, WGA’s AI strategy turnaround provides a structured reset. It combines an independent strategy diagnostic, a readiness rebaseline, portfolio triage to keep, fix or stop initiatives, and ownership and operating model design. Where the gap shows up in oversight, AI governance consulting defines agent roles and decision rights, human oversight, performance targets and monitoring, and audit-ready controls.
Talk to WGA Consulting about an AI Workforce Readiness Assessment
Relevant WGA client experience
A consumer conglomerate had invested heavily in an AI strategy from a top-tier global consulting firm, but the strategy produced no deployable outcomes. It stopped at the slide and never reached the systems, the data or the people who run the business. The group needed working AI capability: agents connected to its systems and data, operating inside its controls and producing measurable results.
WGA is recovering value from the prior work and correcting the direction by identifying measurable sources of business value before selecting technology. It is also building the data, architecture, cybersecurity, governance and organizational capabilities needed to scale AI. At the center is a coordinated portfolio of specialized AI agents, each designed around a defined business role, permissions, workflow and human oversight model, interacting directly with enterprise systems and data within the group’s security and governance controls. WGA was selected after a competitive bidding process, and the engagement is in progress. Read the case study on resetting an enterprise AI strategy.
Sources
- 78% of firms use AI, with US firms expecting to spend $2,000 per employee on AI-related expenditures in 2026, implying around $300 billion in aggregate (nber.org)
- Census Bureau business survey data show that about 18 percent of firms had adopted AI as of year-end 2025 (federalreserve.gov)
- NBER study of corporate executives (ideas.repec.org)




